fix: auto-add unofficial-run precision so ATOM/MTP overlay renders#413
Closed
Oseltamivir wants to merge 1 commit into
Closed
fix: auto-add unofficial-run precision so ATOM/MTP overlay renders#413Oseltamivir wants to merge 1 commit into
Oseltamivir wants to merge 1 commit into
Conversation
When `?unofficialrun=<id>` loads a run whose precision is outside the default `i_prec=fp4`, ScatterGraph silently filtered every overlay point out — the chart shape on the existing run rendered fine, but no overlay was visible. The ATOM/MTP nightly is fp8-only on mi355x, which is why this surfaced now. Model is already auto-switched on overlay load (`computeAutoSwitchDecision`). This adds a sibling `computeAutoAddPrecisionDecision` that augments `selectedPrecisions` with the first run precision when the user's filter and the run's precisions don't intersect. URL-pinned `i_prec` is respected and manual deselects after auto-add aren't undone (same dedupe-ref pattern as the model switch). Verified locally for issue #412 run 26744033059: overlay renders with FP4+FP8 active by default, and `&i_prec=fp4` correctly suppresses the auto-add. Co-authored-by: Bryan Shan <Oseltamivir@users.noreply.github.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #412.
Summary
When
?unofficialrun=<id>loaded a run whose precision was outside the defaulti_prec=fp4,ScatterGraphsilently filtered every overlay point out. The ATOM/MTP nightly run 26744033059 is fp8-only on mi355x, which is why this surfaced now.Added
computeAutoAddPrecisionDecisionto augmentselectedPrecisionswith the first run precision when the user's filter doesn't intersect the overlay's. Mirrors the existingcomputeAutoSwitchDecisionfor model. URL-pinnedi_precis respected.Test plan
pnpm typecheck && pnpm lint && pnpm fmt && pnpm test:unitclean/inference?unofficialRun=26744033059- overlay now renders with FP4+FP8 both active&i_prec=fp4correctly suppresses the auto-add (URL intent respected)🤖 Generated with Claude Code
Note
Low Risk
Client-side filter UX only; mirrors existing unofficial model auto-switch with URL respect and dedupe, covered by unit tests.
Overview
Fixes unofficial-run overlays disappearing when the loaded run’s precision doesn’t match the active filter (e.g. default
fp4while the run isfp8-only), so chart points were filtered out even after model auto-switch.Adds
computeAutoAddPrecisionDecisionalongside the existing model auto-switch helper: when an unofficial run is active for the current model/sequence andselectedPrecisionsdoesn’t overlap the run’s precisions, it picks one run precision (sorted, deterministic) for the UI to merge in. URL-pinnedi_prec, empty overlays, no matching model/seq, and post-deselect dedupe behave like the model path.GlobalFilterContextwires a new effect with a dedupe ref so auto-add runs once per overlay set and doesn’t undo manual precision removal. Unit tests cover the regression and edge cases.Reviewed by Cursor Bugbot for commit 8d6820d. Bugbot is set up for automated code reviews on this repo. Configure here.